4 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
6 * This program is free software; you can redistribute it and/or modify
7 * it under the terms of the GNU General Public License version 2 only,
8 * as published by the Free Software Foundation.
10 * This program is distributed in the hope that it will be useful, but
11 * WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
13 * General Public License version 2 for more details (a copy is included
14 * in the LICENSE file that accompanied this code).
16 * You should have received a copy of the GNU General Public License
17 * version 2 along with this program; If not, see
18 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
20 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
21 * CA 95054 USA or visit www.sun.com if you need additional information or
27 * Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
28 * Use is subject to license terms.
30 * Copyright (c) 2012, 2014, Intel Corporation.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * cl_device and cl_device_type implementation for VVP layer.
38 * Author: Nikita Danilov <nikita.danilov@sun.com>
39 * Author: Jinshan Xiong <jinshan.xiong@intel.com>
42 #define DEBUG_SUBSYSTEM S_LLITE
46 #include "llite_internal.h"
47 #include "vvp_internal.h"
49 /*****************************************************************************
51 * Vvp device and device type functions.
56 * vvp_ prefix stands for "Vfs Vm Posix". It corresponds to historical
57 * "llite_" (var. "ll_") prefix.
60 struct kmem_cache *vvp_lock_kmem;
61 struct kmem_cache *vvp_object_kmem;
62 struct kmem_cache *vvp_req_kmem;
63 static struct kmem_cache *vvp_thread_kmem;
64 static struct kmem_cache *vvp_session_kmem;
65 static struct lu_kmem_descr vvp_caches[] = {
67 .ckd_cache = &vvp_lock_kmem,
68 .ckd_name = "vvp_lock_kmem",
69 .ckd_size = sizeof(struct vvp_lock),
72 .ckd_cache = &vvp_object_kmem,
73 .ckd_name = "vvp_object_kmem",
74 .ckd_size = sizeof(struct vvp_object),
77 .ckd_cache = &vvp_req_kmem,
78 .ckd_name = "vvp_req_kmem",
79 .ckd_size = sizeof(struct vvp_req),
82 .ckd_cache = &vvp_thread_kmem,
83 .ckd_name = "vvp_thread_kmem",
84 .ckd_size = sizeof (struct vvp_thread_info),
87 .ckd_cache = &vvp_session_kmem,
88 .ckd_name = "vvp_session_kmem",
89 .ckd_size = sizeof (struct vvp_session)
96 static void *vvp_key_init(const struct lu_context *ctx,
97 struct lu_context_key *key)
99 struct vvp_thread_info *info;
101 OBD_SLAB_ALLOC_PTR_GFP(info, vvp_thread_kmem, GFP_NOFS);
103 info = ERR_PTR(-ENOMEM);
107 static void vvp_key_fini(const struct lu_context *ctx,
108 struct lu_context_key *key, void *data)
110 struct vvp_thread_info *info = data;
111 OBD_SLAB_FREE_PTR(info, vvp_thread_kmem);
114 static void *vvp_session_key_init(const struct lu_context *ctx,
115 struct lu_context_key *key)
117 struct vvp_session *session;
119 OBD_SLAB_ALLOC_PTR_GFP(session, vvp_session_kmem, GFP_NOFS);
121 session = ERR_PTR(-ENOMEM);
125 static void vvp_session_key_fini(const struct lu_context *ctx,
126 struct lu_context_key *key, void *data)
128 struct vvp_session *session = data;
129 OBD_SLAB_FREE_PTR(session, vvp_session_kmem);
133 struct lu_context_key vvp_key = {
134 .lct_tags = LCT_CL_THREAD,
135 .lct_init = vvp_key_init,
136 .lct_fini = vvp_key_fini
139 struct lu_context_key vvp_session_key = {
140 .lct_tags = LCT_SESSION,
141 .lct_init = vvp_session_key_init,
142 .lct_fini = vvp_session_key_fini
145 /* type constructor/destructor: vvp_type_{init,fini,start,stop}(). */
146 LU_TYPE_INIT_FINI(vvp, &ccc_key, &vvp_key, &vvp_session_key);
148 static const struct lu_device_operations vvp_lu_ops = {
149 .ldo_object_alloc = vvp_object_alloc
152 static const struct cl_device_operations vvp_cl_ops = {
153 .cdo_req_init = vvp_req_init,
156 static struct lu_device *vvp_device_free(const struct lu_env *env,
159 struct vvp_device *vdv = lu2vvp_dev(d);
160 struct cl_site *site = lu2cl_site(d->ld_site);
161 struct lu_device *next = cl2lu_dev(vdv->vdv_next);
163 if (d->ld_site != NULL) {
168 cl_device_fini(lu2cl_dev(d));
173 static struct lu_device *vvp_device_alloc(const struct lu_env *env,
174 struct lu_device_type *t,
175 struct lustre_cfg *cfg)
177 struct vvp_device *vdv;
178 struct lu_device *lud;
179 struct cl_site *site;
185 RETURN(ERR_PTR(-ENOMEM));
187 lud = &vdv->vdv_cl.cd_lu_dev;
188 cl_device_init(&vdv->vdv_cl, t);
189 vvp2lu_dev(vdv)->ld_ops = &vvp_lu_ops;
190 vdv->vdv_cl.cd_ops = &vvp_cl_ops;
194 rc = cl_site_init(site, &vdv->vdv_cl);
196 rc = lu_site_init_finish(&site->cs_lu);
198 LASSERT(lud->ld_site == NULL);
199 CERROR("Cannot init lu_site, rc %d.\n", rc);
205 vvp_device_free(env, lud);
211 static int vvp_device_init(const struct lu_env *env, struct lu_device *d,
212 const char *name, struct lu_device *next)
214 struct vvp_device *vdv;
219 vdv->vdv_next = lu2cl_dev(next);
221 LASSERT(d->ld_site != NULL && next->ld_type != NULL);
222 next->ld_site = d->ld_site;
223 rc = next->ld_type->ldt_ops->ldto_device_init(
224 env, next, next->ld_type->ldt_name, NULL);
227 lu_ref_add(&next->ld_reference, "lu-stack", &lu_site_init);
232 static struct lu_device *vvp_device_fini(const struct lu_env *env,
235 return cl2lu_dev(lu2vvp_dev(d)->vdv_next);
238 static const struct lu_device_type_operations vvp_device_type_ops = {
239 .ldto_init = vvp_type_init,
240 .ldto_fini = vvp_type_fini,
242 .ldto_start = vvp_type_start,
243 .ldto_stop = vvp_type_stop,
245 .ldto_device_alloc = vvp_device_alloc,
246 .ldto_device_free = vvp_device_free,
247 .ldto_device_init = vvp_device_init,
248 .ldto_device_fini = vvp_device_fini,
251 struct lu_device_type vvp_device_type = {
252 .ldt_tags = LU_DEVICE_CL,
253 .ldt_name = LUSTRE_VVP_NAME,
254 .ldt_ops = &vvp_device_type_ops,
255 .ldt_ctx_tags = LCT_CL_THREAD
259 * A mutex serializing calls to vvp_inode_fini() under extreme memory
260 * pressure, when environments cannot be allocated.
262 int vvp_global_init(void)
266 result = lu_kmem_init(vvp_caches);
268 result = ccc_global_init(&vvp_device_type);
270 lu_kmem_fini(vvp_caches);
275 void vvp_global_fini(void)
277 ccc_global_fini(&vvp_device_type);
278 lu_kmem_fini(vvp_caches);
282 /*****************************************************************************
284 * mirror obd-devices into cl devices.
288 int cl_sb_init(struct super_block *sb)
290 struct ll_sb_info *sbi;
291 struct cl_device *cl;
297 env = cl_env_get(&refcheck);
299 cl = cl_type_setup(env, NULL, &vvp_device_type,
300 sbi->ll_dt_exp->exp_obd->obd_lu_dev);
302 cl2vvp_dev(cl)->vdv_sb = sb;
304 sbi->ll_site = cl2lu_dev(cl)->ld_site;
306 cl_env_put(env, &refcheck);
312 int cl_sb_fini(struct super_block *sb)
314 struct ll_sb_info *sbi;
316 struct cl_device *cld;
322 env = cl_env_get(&refcheck);
327 cl_stack_fini(env, cld);
331 cl_env_put(env, &refcheck);
334 CERROR("Cannot cleanup cl-stack due to memory shortage.\n");
335 result = PTR_ERR(env);
341 /****************************************************************************
343 * /proc/fs/lustre/llite/$MNT/dump_page_cache
345 ****************************************************************************/
348 * To represent contents of a page cache as a byte stream, following
349 * information if encoded in 64bit offset:
351 * - file hash bucket in lu_site::ls_hash[] 28bits
353 * - how far file is from bucket head 4bits
355 * - page index 32bits
357 * First two data identify a file in the cache uniquely.
360 #define PGC_OBJ_SHIFT (32 + 4)
361 #define PGC_DEPTH_SHIFT (32)
363 struct vvp_pgcache_id {
369 struct lu_object_header *vpi_obj;
372 static void vvp_pgcache_id_unpack(loff_t pos, struct vvp_pgcache_id *id)
374 CLASSERT(sizeof(pos) == sizeof(__u64));
376 id->vpi_index = pos & 0xffffffff;
377 id->vpi_depth = (pos >> PGC_DEPTH_SHIFT) & 0xf;
378 id->vpi_bucket = ((unsigned long long)pos >> PGC_OBJ_SHIFT);
381 static loff_t vvp_pgcache_id_pack(struct vvp_pgcache_id *id)
384 ((__u64)id->vpi_index) |
385 ((__u64)id->vpi_depth << PGC_DEPTH_SHIFT) |
386 ((__u64)id->vpi_bucket << PGC_OBJ_SHIFT);
389 static int vvp_pgcache_obj_get(cfs_hash_t *hs, cfs_hash_bd_t *bd,
390 struct hlist_node *hnode, void *data)
392 struct vvp_pgcache_id *id = data;
393 struct lu_object_header *hdr = cfs_hash_object(hs, hnode);
395 if (id->vpi_curdep-- > 0)
396 return 0; /* continue */
398 if (lu_object_is_dying(hdr))
401 cfs_hash_get(hs, hnode);
406 static struct cl_object *vvp_pgcache_obj(const struct lu_env *env,
407 struct lu_device *dev,
408 struct vvp_pgcache_id *id)
410 LASSERT(lu_device_is_cl(dev));
412 id->vpi_depth &= 0xf;
414 id->vpi_curdep = id->vpi_depth;
416 cfs_hash_hlist_for_each(dev->ld_site->ls_obj_hash, id->vpi_bucket,
417 vvp_pgcache_obj_get, id);
418 if (id->vpi_obj != NULL) {
419 struct lu_object *lu_obj;
421 lu_obj = lu_object_locate(id->vpi_obj, dev->ld_type);
422 if (lu_obj != NULL) {
423 lu_object_ref_add(lu_obj, "dump", current);
424 return lu2cl(lu_obj);
426 lu_object_put(env, lu_object_top(id->vpi_obj));
428 } else if (id->vpi_curdep > 0) {
434 static loff_t vvp_pgcache_find(const struct lu_env *env,
435 struct lu_device *dev, loff_t pos)
437 struct cl_object *clob;
438 struct lu_site *site;
439 struct vvp_pgcache_id id;
442 vvp_pgcache_id_unpack(pos, &id);
445 if (id.vpi_bucket >= CFS_HASH_NHLIST(site->ls_obj_hash))
447 clob = vvp_pgcache_obj(env, dev, &id);
449 struct inode *inode = vvp_object_inode(clob);
453 nr = find_get_pages_contig(inode->i_mapping,
454 id.vpi_index, 1, &vmpage);
456 id.vpi_index = vmpage->index;
457 /* Cant support over 16T file */
458 nr = !(vmpage->index > 0xffffffff);
459 page_cache_release(vmpage);
462 lu_object_ref_del(&clob->co_lu, "dump", current);
463 cl_object_put(env, clob);
465 return vvp_pgcache_id_pack(&id);
467 /* to the next object. */
470 if (id.vpi_depth == 0 && ++id.vpi_bucket == 0)
476 #define seq_page_flag(seq, page, flag, has_flags) do { \
477 if (test_bit(PG_##flag, &(page)->flags)) { \
478 seq_printf(seq, "%s"#flag, has_flags ? "|" : ""); \
483 static void vvp_pgcache_page_show(const struct lu_env *env,
484 struct seq_file *seq, struct cl_page *page)
486 struct vvp_page *vpg;
490 vpg = cl2vvp_page(cl_page_at(page, &vvp_device_type));
491 vmpage = vpg->vpg_page;
492 seq_printf(seq, " %5i | %p %p %s %s %s %s | %p "DFID"(%p) %lu %u [",
496 vpg->vpg_write_queued ? "wq" : "- ",
497 vpg->vpg_defer_uptodate ? "du" : "- ",
498 PageWriteback(vmpage) ? "wb" : "-",
500 PFID(ll_inode2fid(vmpage->mapping->host)),
501 vmpage->mapping->host, vmpage->index,
504 seq_page_flag(seq, vmpage, locked, has_flags);
505 seq_page_flag(seq, vmpage, error, has_flags);
506 seq_page_flag(seq, vmpage, referenced, has_flags);
507 seq_page_flag(seq, vmpage, uptodate, has_flags);
508 seq_page_flag(seq, vmpage, dirty, has_flags);
509 seq_page_flag(seq, vmpage, writeback, has_flags);
510 seq_printf(seq, "%s]\n", has_flags ? "" : "-");
513 static int vvp_pgcache_show(struct seq_file *f, void *v)
516 struct ll_sb_info *sbi;
517 struct cl_object *clob;
519 struct vvp_pgcache_id id;
523 env = cl_env_get(&refcheck);
526 vvp_pgcache_id_unpack(pos, &id);
528 clob = vvp_pgcache_obj(env, &sbi->ll_cl->cd_lu_dev, &id);
530 struct inode *inode = vvp_object_inode(clob);
531 struct cl_page *page = NULL;
534 result = find_get_pages_contig(inode->i_mapping,
535 id.vpi_index, 1, &vmpage);
538 page = cl_vmpage_page(vmpage, clob);
541 page_cache_release(vmpage);
544 seq_printf(f, "%8x@"DFID": ", id.vpi_index,
545 PFID(lu_object_fid(&clob->co_lu)));
547 vvp_pgcache_page_show(env, f, page);
548 cl_page_put(env, page);
550 seq_puts(f, "missing\n");
551 lu_object_ref_del(&clob->co_lu, "dump", current);
552 cl_object_put(env, clob);
554 seq_printf(f, "%llx missing\n", pos);
555 cl_env_put(env, &refcheck);
558 result = PTR_ERR(env);
562 static void *vvp_pgcache_start(struct seq_file *f, loff_t *pos)
564 struct ll_sb_info *sbi;
570 env = cl_env_get(&refcheck);
573 if (sbi->ll_site->ls_obj_hash->hs_cur_bits > 64 - PGC_OBJ_SHIFT)
574 pos = ERR_PTR(-EFBIG);
576 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev,
581 cl_env_put(env, &refcheck);
586 static void *vvp_pgcache_next(struct seq_file *f, void *v, loff_t *pos)
588 struct ll_sb_info *sbi;
592 env = cl_env_get(&refcheck);
595 *pos = vvp_pgcache_find(env, &sbi->ll_cl->cd_lu_dev, *pos + 1);
598 cl_env_put(env, &refcheck);
603 static void vvp_pgcache_stop(struct seq_file *f, void *v)
608 static struct seq_operations vvp_pgcache_ops = {
609 .start = vvp_pgcache_start,
610 .next = vvp_pgcache_next,
611 .stop = vvp_pgcache_stop,
612 .show = vvp_pgcache_show
615 static int vvp_dump_pgcache_seq_open(struct inode *inode, struct file *filp)
617 struct ll_sb_info *sbi = PDE_DATA(inode);
618 struct seq_file *seq;
621 result = seq_open(filp, &vvp_pgcache_ops);
623 seq = filp->private_data;
629 const struct file_operations vvp_dump_pgcache_file_ops = {
630 .owner = THIS_MODULE,
631 .open = vvp_dump_pgcache_seq_open,
634 .release = seq_release,